yamle.methods.rbnn module#

yamle.methods.rbnn.replace_with_rbnn(model, num_members, prior_mean, log_variance, prior_log_variance, method)[source]#
This method is used to replace all the nn.Linear, nn.Conv2d layers

with a LinearRBNN and Conv2dRBNN respectively.

Parameters:

  • model (nn.Module) – The model to replace the layers in.

  • num_members (int) – The number of members in the ensemble.

  • prior_mean (float) – The mean of the prior distribution.

  • log_variance (float) – The initial value of the log of the standard deviation of the weights.

  • prior_log_variance (float) – The initial value of the log of the standard deviation of the prior distribution.

  • method (str) – The method whether additive or multiplicative to be used for the rank-1 approximation.

Return type:

Module

class yamle.methods.rbnn.RBNNMethod(prior_mean=1.0, log_variance=-3.0, prior_log_variance=-3.0, method='additive', **kwargs)[source]#

Bases: SVIMethod

This class is the extension of the base method for Rank-1 Bayesian Neural Networks.

Parameters:

  • num_members (int) – The number of members in the ensemble.

  • prior_mean (float) – The mean of the prior distribution.

  • log_variance (float) – The initial value of the log of the standard deviation of the weights.

  • prior_log_variance (float) – The initial value of the log of the standard deviation of the prior distribution.

  • method (str) – The method whether additive or multiplicative to be used for the rank-1 approximation.

static add_specific_args(parent_parser)[source]#

This method adds the specific arguments for the MIMO method.

Return type:

ArgumentParser

test_name: Optional[str]#
prepare_data_per_node: bool#
allow_zero_length_dataloader_with_multiple_devices: bool#
training: bool#